|
0
|
1 /* Test and benchmark of a couple of parallel sorting algorithms.
|
|
|
2 Copyright (C) 2008 Free Software Foundation, Inc.
|
|
|
3
|
|
|
4 GCC is free software; you can redistribute it and/or modify it under
|
|
|
5 the terms of the GNU General Public License as published by the Free
|
|
|
6 Software Foundation; either version 3, or (at your option) any later
|
|
|
7 version.
|
|
|
8
|
|
|
9 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
|
|
|
10 WARRANTY; without even the implied warranty of MERCHANTABILITY or
|
|
|
11 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
|
12 for more details.
|
|
|
13
|
|
|
14 You should have received a copy of the GNU General Public License
|
|
|
15 along with GCC; see the file COPYING3. If not see
|
|
|
16 <http://www.gnu.org/licenses/>. */
|
|
|
17
|
|
|
18 #include <limits.h>
|
|
|
19 #include <omp.h>
|
|
|
20 #include <stdbool.h>
|
|
|
21 #include <stdio.h>
|
|
|
22 #include <stdlib.h>
|
|
|
23 #include <string.h>
|
|
|
24
|
|
|
25 int failures;
|
|
|
26
|
|
|
27 #define THRESHOLD 100
|
|
|
28
|
|
|
29 static void
|
|
|
30 verify (const char *name, double stime, int *array, int count)
|
|
|
31 {
|
|
|
32 int i;
|
|
|
33 double etime = omp_get_wtime ();
|
|
|
34
|
|
|
35 printf ("%s: %g\n", name, etime - stime);
|
|
|
36 for (i = 1; i < count; i++)
|
|
|
37 if (array[i] < array[i - 1])
|
|
|
38 {
|
|
|
39 printf ("%s: incorrectly sorted\n", name);
|
|
|
40 failures = 1;
|
|
|
41 }
|
|
|
42 }
|
|
|
43
|
|
|
44 static void
|
|
|
45 insertsort (int *array, int s, int e)
|
|
|
46 {
|
|
|
47 int i, j, val;
|
|
|
48 for (i = s + 1; i <= e; i++)
|
|
|
49 {
|
|
|
50 val = array[i];
|
|
|
51 j = i;
|
|
|
52 while (j-- > s && val < array[j])
|
|
|
53 array[j + 1] = array[j];
|
|
|
54 array[j + 1] = val;
|
|
|
55 }
|
|
|
56 }
|
|
|
57
|
|
|
58 struct int_pair
|
|
|
59 {
|
|
|
60 int lo;
|
|
|
61 int hi;
|
|
|
62 };
|
|
|
63
|
|
|
64 struct int_pair_stack
|
|
|
65 {
|
|
|
66 struct int_pair *top;
|
|
|
67 #define STACK_SIZE 4 * CHAR_BIT * sizeof (int)
|
|
|
68 struct int_pair arr[STACK_SIZE];
|
|
|
69 };
|
|
|
70
|
|
|
71 static inline void
|
|
|
72 init_int_pair_stack (struct int_pair_stack *stack)
|
|
|
73 {
|
|
|
74 stack->top = &stack->arr[0];
|
|
|
75 }
|
|
|
76
|
|
|
77 static inline void
|
|
|
78 push_int_pair_stack (struct int_pair_stack *stack, int lo, int hi)
|
|
|
79 {
|
|
|
80 stack->top->lo = lo;
|
|
|
81 stack->top->hi = hi;
|
|
|
82 stack->top++;
|
|
|
83 }
|
|
|
84
|
|
|
85 static inline void
|
|
|
86 pop_int_pair_stack (struct int_pair_stack *stack, int *lo, int *hi)
|
|
|
87 {
|
|
|
88 stack->top--;
|
|
|
89 *lo = stack->top->lo;
|
|
|
90 *hi = stack->top->hi;
|
|
|
91 }
|
|
|
92
|
|
|
93 static inline int
|
|
|
94 size_int_pair_stack (struct int_pair_stack *stack)
|
|
|
95 {
|
|
|
96 return stack->top - &stack->arr[0];
|
|
|
97 }
|
|
|
98
|
|
|
99 static inline void
|
|
|
100 busy_wait (void)
|
|
|
101 {
|
|
|
102 #if defined __i386__ || defined __x86_64__
|
|
|
103 __asm volatile ("rep; nop" : : : "memory");
|
|
|
104 #elif defined __ia64__
|
|
|
105 __asm volatile ("hint @pause" : : : "memory");
|
|
|
106 #elif defined __sparc__ && (defined __arch64__ || defined __sparc_v9__)
|
|
|
107 __asm volatile ("membar #LoadLoad" : : : "memory");
|
|
|
108 #else
|
|
|
109 __asm volatile ("" : : : "memory");
|
|
|
110 #endif
|
|
|
111 }
|
|
|
112
|
|
|
113 static inline void
|
|
|
114 swap (int *array, int a, int b)
|
|
|
115 {
|
|
|
116 int val = array[a];
|
|
|
117 array[a] = array[b];
|
|
|
118 array[b] = val;
|
|
|
119 }
|
|
|
120
|
|
|
121 static inline int
|
|
|
122 choose_pivot (int *array, int lo, int hi)
|
|
|
123 {
|
|
|
124 int mid = (lo + hi) / 2;
|
|
|
125
|
|
|
126 if (array[mid] < array[lo])
|
|
|
127 swap (array, lo, mid);
|
|
|
128 if (array[hi] < array[mid])
|
|
|
129 {
|
|
|
130 swap (array, mid, hi);
|
|
|
131 if (array[mid] < array[lo])
|
|
|
132 swap (array, lo, mid);
|
|
|
133 }
|
|
|
134 return array[mid];
|
|
|
135 }
|
|
|
136
|
|
|
137 static inline int
|
|
|
138 partition (int *array, int lo, int hi)
|
|
|
139 {
|
|
|
140 int pivot = choose_pivot (array, lo, hi);
|
|
|
141 int left = lo;
|
|
|
142 int right = hi;
|
|
|
143
|
|
|
144 for (;;)
|
|
|
145 {
|
|
|
146 while (array[++left] < pivot);
|
|
|
147 while (array[--right] > pivot);
|
|
|
148 if (left >= right)
|
|
|
149 break;
|
|
|
150 swap (array, left, right);
|
|
|
151 }
|
|
|
152 return left;
|
|
|
153 }
|
|
|
154
|
|
|
155 static void
|
|
|
156 sort1 (int *array, int count)
|
|
|
157 {
|
|
|
158 omp_lock_t lock;
|
|
|
159 struct int_pair_stack global_stack;
|
|
|
160 int busy = 1;
|
|
|
161 int num_threads;
|
|
|
162
|
|
|
163 omp_init_lock (&lock);
|
|
|
164 init_int_pair_stack (&global_stack);
|
|
|
165 #pragma omp parallel firstprivate (array, count)
|
|
|
166 {
|
|
|
167 int lo = 0, hi = 0, mid, next_lo, next_hi;
|
|
|
168 bool idle = true;
|
|
|
169 struct int_pair_stack local_stack;
|
|
|
170
|
|
|
171 init_int_pair_stack (&local_stack);
|
|
|
172 if (omp_get_thread_num () == 0)
|
|
|
173 {
|
|
|
174 num_threads = omp_get_num_threads ();
|
|
|
175 hi = count - 1;
|
|
|
176 idle = false;
|
|
|
177 }
|
|
|
178
|
|
|
179 for (;;)
|
|
|
180 {
|
|
|
181 if (hi - lo < THRESHOLD)
|
|
|
182 {
|
|
|
183 insertsort (array, lo, hi);
|
|
|
184 lo = hi;
|
|
|
185 }
|
|
|
186 if (lo >= hi)
|
|
|
187 {
|
|
|
188 if (size_int_pair_stack (&local_stack) == 0)
|
|
|
189 {
|
|
|
190 again:
|
|
|
191 omp_set_lock (&lock);
|
|
|
192 if (size_int_pair_stack (&global_stack) == 0)
|
|
|
193 {
|
|
|
194 if (!idle)
|
|
|
195 busy--;
|
|
|
196 if (busy == 0)
|
|
|
197 {
|
|
|
198 omp_unset_lock (&lock);
|
|
|
199 break;
|
|
|
200 }
|
|
|
201 omp_unset_lock (&lock);
|
|
|
202 idle = true;
|
|
|
203 while (size_int_pair_stack (&global_stack) == 0
|
|
|
204 && busy)
|
|
|
205 busy_wait ();
|
|
|
206 goto again;
|
|
|
207 }
|
|
|
208 if (idle)
|
|
|
209 busy++;
|
|
|
210 pop_int_pair_stack (&global_stack, &lo, &hi);
|
|
|
211 omp_unset_lock (&lock);
|
|
|
212 idle = false;
|
|
|
213 }
|
|
|
214 else
|
|
|
215 pop_int_pair_stack (&local_stack, &lo, &hi);
|
|
|
216 }
|
|
|
217
|
|
|
218 mid = partition (array, lo, hi);
|
|
|
219 if (mid - lo < hi - mid)
|
|
|
220 {
|
|
|
221 next_lo = mid;
|
|
|
222 next_hi = hi;
|
|
|
223 hi = mid - 1;
|
|
|
224 }
|
|
|
225 else
|
|
|
226 {
|
|
|
227 next_lo = lo;
|
|
|
228 next_hi = mid - 1;
|
|
|
229 lo = mid;
|
|
|
230 }
|
|
|
231
|
|
|
232 if (next_hi - next_lo < THRESHOLD)
|
|
|
233 insertsort (array, next_lo, next_hi);
|
|
|
234 else
|
|
|
235 {
|
|
|
236 if (size_int_pair_stack (&global_stack) < num_threads - 1)
|
|
|
237 {
|
|
|
238 int size;
|
|
|
239
|
|
|
240 omp_set_lock (&lock);
|
|
|
241 size = size_int_pair_stack (&global_stack);
|
|
|
242 if (size < num_threads - 1 && size < STACK_SIZE)
|
|
|
243 push_int_pair_stack (&global_stack, next_lo, next_hi);
|
|
|
244 else
|
|
|
245 push_int_pair_stack (&local_stack, next_lo, next_hi);
|
|
|
246 omp_unset_lock (&lock);
|
|
|
247 }
|
|
|
248 else
|
|
|
249 push_int_pair_stack (&local_stack, next_lo, next_hi);
|
|
|
250 }
|
|
|
251 }
|
|
|
252 }
|
|
|
253 omp_destroy_lock (&lock);
|
|
|
254 }
|
|
|
255
|
|
|
256 static void
|
|
|
257 sort2_1 (int *array, int lo, int hi, int num_threads, int *busy)
|
|
|
258 {
|
|
|
259 int mid;
|
|
|
260
|
|
|
261 if (hi - lo < THRESHOLD)
|
|
|
262 {
|
|
|
263 insertsort (array, lo, hi);
|
|
|
264 return;
|
|
|
265 }
|
|
|
266
|
|
|
267 mid = partition (array, lo, hi);
|
|
|
268
|
|
|
269 if (*busy >= num_threads)
|
|
|
270 {
|
|
|
271 sort2_1 (array, lo, mid - 1, num_threads, busy);
|
|
|
272 sort2_1 (array, mid, hi, num_threads, busy);
|
|
|
273 return;
|
|
|
274 }
|
|
|
275
|
|
|
276 #pragma omp atomic
|
|
|
277 *busy += 1;
|
|
|
278
|
|
|
279 #pragma omp parallel num_threads (2) \
|
|
|
280 firstprivate (array, lo, hi, mid, num_threads, busy)
|
|
|
281 {
|
|
|
282 if (omp_get_thread_num () == 0)
|
|
|
283 sort2_1 (array, lo, mid - 1, num_threads, busy);
|
|
|
284 else
|
|
|
285 {
|
|
|
286 sort2_1 (array, mid, hi, num_threads, busy);
|
|
|
287 #pragma omp atomic
|
|
|
288 *busy -= 1;
|
|
|
289 }
|
|
|
290 }
|
|
|
291 }
|
|
|
292
|
|
|
293 static void
|
|
|
294 sort2 (int *array, int count)
|
|
|
295 {
|
|
|
296 int num_threads;
|
|
|
297 int busy = 1;
|
|
|
298
|
|
|
299 #pragma omp parallel
|
|
|
300 #pragma omp single nowait
|
|
|
301 num_threads = omp_get_num_threads ();
|
|
|
302
|
|
|
303 sort2_1 (array, 0, count - 1, num_threads, &busy);
|
|
|
304 }
|
|
|
305
|
|
|
306 #if _OPENMP >= 200805
|
|
|
307 static void
|
|
|
308 sort3_1 (int *array, int lo, int hi)
|
|
|
309 {
|
|
|
310 int mid;
|
|
|
311
|
|
|
312 if (hi - lo < THRESHOLD)
|
|
|
313 {
|
|
|
314 insertsort (array, lo, hi);
|
|
|
315 return;
|
|
|
316 }
|
|
|
317
|
|
|
318 mid = partition (array, lo, hi);
|
|
|
319 #pragma omp task
|
|
|
320 sort3_1 (array, lo, mid - 1);
|
|
|
321 sort3_1 (array, mid, hi);
|
|
|
322 }
|
|
|
323
|
|
|
324 static void
|
|
|
325 sort3 (int *array, int count)
|
|
|
326 {
|
|
|
327 #pragma omp parallel
|
|
|
328 #pragma omp single
|
|
|
329 sort3_1 (array, 0, count - 1);
|
|
|
330 }
|
|
|
331 #endif
|
|
|
332
|
|
|
333 int
|
|
|
334 main (int argc, char **argv)
|
|
|
335 {
|
|
|
336 int i, count = 1000000;
|
|
|
337 double stime;
|
|
|
338 int *unsorted, *sorted, num_threads;
|
|
|
339 if (argc >= 2)
|
|
|
340 count = strtoul (argv[1], NULL, 0);
|
|
|
341
|
|
|
342 unsorted = malloc (count * sizeof (int));
|
|
|
343 sorted = malloc (count * sizeof (int));
|
|
|
344 if (unsorted == NULL || sorted == NULL)
|
|
|
345 {
|
|
|
346 puts ("allocation failure");
|
|
|
347 exit (1);
|
|
|
348 }
|
|
|
349
|
|
|
350 srand (0xdeadbeef);
|
|
|
351 for (i = 0; i < count; i++)
|
|
|
352 unsorted[i] = rand ();
|
|
|
353
|
|
|
354 omp_set_nested (1);
|
|
|
355 omp_set_dynamic (0);
|
|
|
356 #pragma omp parallel
|
|
|
357 #pragma omp single nowait
|
|
|
358 num_threads = omp_get_num_threads ();
|
|
|
359 printf ("Threads: %d\n", num_threads);
|
|
|
360
|
|
|
361 memcpy (sorted, unsorted, count * sizeof (int));
|
|
|
362 stime = omp_get_wtime ();
|
|
|
363 sort1 (sorted, count);
|
|
|
364 verify ("sort1", stime, sorted, count);
|
|
|
365
|
|
|
366 memcpy (sorted, unsorted, count * sizeof (int));
|
|
|
367 stime = omp_get_wtime ();
|
|
|
368 sort2 (sorted, count);
|
|
|
369 verify ("sort2", stime, sorted, count);
|
|
|
370
|
|
|
371 #if _OPENMP >= 200805
|
|
|
372 memcpy (sorted, unsorted, count * sizeof (int));
|
|
|
373 stime = omp_get_wtime ();
|
|
|
374 sort3 (sorted, count);
|
|
|
375 verify ("sort3", stime, sorted, count);
|
|
|
376 #endif
|
|
|
377
|
|
|
378 return 0;
|
|
|
379 }
|
